Book launch party at SEA: Pretrained Transformers for Text Ranking

***
IMPORTANT: You will be able to view the Zoom link once you 'attend' the meetup on this page.
***
The last edition of SEA this year is a special one! We are thrilled to announce that this edition is the book launch party for the book: Pretrained Transformers for Text Ranking: BERT and Beyond, by Jimmy Lin, Rodrigo Nogueira, and Andrew Yates.
To celebrate, we've organized a book giveaway and invited three distinguished speakers from the community to share their fun stories about the book:

Chenyan Xiong - inventor of the ANCE dense retrieval model!
Ellen Voorhees - who runs TREC!
Kyunghyun Cho - who started the "BERT for IR" craze with Rodrigo Nogueira!

Please join us for this special event! Attendees will also have a chance to win a free copy of the book (to be delivered by mail).

17:00-17:10: Introduction by Andrew Yates (University of Amsterdam)

17:15-17:30: Chenyan Xiong (Microsoft Research AI)

Title: Simplify Complex QA System with Embedding Dot-Products
Abstract: Complex question answering systems aim to answer questions that require some terms of “complex reasoning” to combine evidence from multiple documents. Previously, the go-to solution is to leverage structured semantics from knowledge graphs and conduct the reasoning as traverses in the graph. With dense retrieval, we now can simply perform multiple embedding dot product operations in the learned representation space to achieve the same reasoning capability, with all things embedded in the learned embedding space. In this talk, I will show how this complex QA task is done in my previous research, the complicated systems we built, and how it can be done in the new representation-centric operation, where we not only get rid of the requirement of structured data, but also no longer need intermediate annotations at evidence level.

17:30-17:45: Ellen Voorhees (National Institute of Standards and Technology)

Title: Everything Old is New Again

Abstract: The first TREC was held in 1992 and was motivated in part by the desire to "increase the availability of appropriate evaluation techniques...including the development of new evaluation techniques more applicable to current systems". Thirty years later, TREC contains the Deep Learning track that is once again motivated by the desire to develop appropriate evaluation techniques for today's current systems, those based on dense network models. This talk will take a brief look at how retrieval system developments necessitate changes to evaluation infrastructure.

17:45-18:00: Kyunghyun Cho (New York University)

Title: How BERT came to information retrieval
Abstract: Rodrigo Nogueira was a Ph.D. student who had been working on fingerprint analysis when I arrived at NYU in Fall 2015. Four years later in 2019, Rodrigo and I posted a 3-pages-long paper on arXiv describing how to use BERT for query-based passage re-ranking. In this talk, I will tell you my experience seeing how Rodrigo's research evolved from fingerprint analysis to information retrieval over those four years.

---
This is (still) an online event. The URL will be shared close to the day of the event. All times are Amsterdam times.

Keeping track? These SEA talks are SEA talks #183, #184, #185, and #186.